In February, McKinsey Global Institute predicted that 45 million Americans, or a quarter of the workforce, would lose their jobs because of automation by 2030. This was up from 2017 estimate that 39 million would be automated without work, due to the economic dislocation of COVID-19. Historically, companies have tended to replace some of the workers they lay off during recessions with machines.
The fear of mass unemployment caused by robots has become increasingly common. Andrew Yang, who currently leads the polls for the Democratic nomination for New York’s next mayor, has made him a mainstay of his unorthodox 2020 presidential campaign. Lack of jobs ahead, Yang noted, justified giving all Americans a monthly government income of $ 1,000.
But take a close look at studies predicting job losses due to automation, and you’ll find less reason to worry (although there is still reason to consider a universal basic income). Robots are mostly not coming, at least not anytime soon.
For starters, there is a huge difference between “robots” and “automation”. In the past, many elevators were operated by people. An office building in DC still had elevator operators in the 2010s. They haven’t been replaced by humanoid robots that listen to riders’ demands and manipulate a lever with mechanical fingers. They have been replaced by a row of buttons that the riders press themselves. A lot of automation works that way.
The distinction is important because automation happens all the time. Over the past 150 years, we have gone from a nation of farmers to a nation of factory workers to a nation of white collar and service workers, with much of this momentous change driven by l ‘automating. But while regional economies have been disrupted and recessions have created periodic unemployment crises, there has never been a chronic, structural shortage of jobs nationwide. New inventions create new markets and the jobs that go with them.
The robot work apocalypse scenario is based on the assumption that the next wave of automation technology will be fundamentally different. Artificial intelligence in particular is believed to be advancing so rapidly that replacement jobs are not keeping pace. People wonder if our fragile and imperfect species will be needed any longer.
But that’s not what forecasters say. The boom in predicting robot job losses kicked into high gear in 2013, when a pair of researchers from the University of Oxford valued that 47 percent of American jobs are “at risk” of being computerized. The report has been widely cited, including in White House Reports.
To arrive at this estimate, a team of machine learning experts looked at 70 trades, each of which had been analyzed by the US Department of Labor and broken down into dozens of distinct tasks and skills. Experts looked at each task and made an educated guess as to whether it could be automated, assuming cutting edge technology, the huge datasets that power modern AI, and future technical breakthroughs that haven’t yet happened. produced. They used those estimates to write an algorithm that automatically analyzed hundreds of other jobs.
“At risk” of automation does not mean, in this analysis, “likely to be automated”. This means that “could theoretically be automated if someone had unlimited time, money and access to the latest AI.” It’s a huge difference. Maybe the engineers at Boston Dynamics, who make these viral videos ominous humanoid robots, could spend millions of dollars to build a robotic version of the guy standing around the corner spinning the big pointed sign that says “Woing Out of Business Sale !!!” But they won’t, because no one would buy this robot, because they can just hire the guy for $ 10 an hour.
The recent McKinsey report takes this into account, estimating the cost of developing a new automation technology, the cost of labor it would replace, and the time it would take for widespread adoption. That’s why his estimate is 27 percent of jobs, not 47 percent. But here, too, definitions matter.
McKinsey predicts that of the 49.1 million who will see their jobs displaced by automation, 32 million will remain in the same occupation and 2.2 million will remain in the same occupational category. The number of people who will be to lose their jobs in the sense “must find a new line of work” is only 14.9 million. Not 27 percent, but 9 percent.
This is because automation is more likely to change jobs than destroy them. Machines will perform an increasing share of boring and routine tasks, and people will shift to more humane jobs. When hundreds of thousands of ATMs were deployed in the 1980s and 1990s, the number of bank tellers increased, not decreased, as lower labor costs allowed banks to open more banks. branches. Now the machines are counting the money and people are selling you car loans. Automation works especially well when workers are partners in shaping their new relationships with machines.
Nine percent of jobs is still a lot. But the optimal number is not zero. The White House Automation Report notes that about 6% of jobs in the US economy are cut every three months through the normal process of downsizing or shutting down some businesses while others start up and grow. .
Job loss due to automation does exist. In 2020, economists Daron Acemoglu and Pascual Restrepo find that every new industrial robot deployed in the United States between 1990 and 2007 replaced 3.3 workers, even after accounting for the positive economic effects of more productive firms. It was a small impact — one in 1,000 workers — but very real.
The question of who is replaced is also difficult. The automation of the 19th century often replaced the better paid skilled craftsmen. The automation of the twenty-first century hits the lowest-paid and low-skilled workers the hardest. Recent recessions have been brutal for working class families, who often never regain lost economic ground. America’s unemployment insurance system is creaky, inadequate, and in dire need of reform.
The wildest robotic dystopia scenarios often proceed from metaphor failures. Many powerful new AI systems use methods called “neural networks”, which people assume to mean “like a human brain”. They are not like human brains. AI is pattern recognition. Alexa knows that some spoken sounds correspond to the letter sequence “peanut butter”, which is remarkable. But Alexa has no idea what “peanut butter” means or why it tastes good with jelly.
The most sober predictions of job loss in automation always rely on a network of nested predictions that may not come true. Five years ago, it looked like we were about to see robot taxis and freight trucks go mainstream. Today we are stuck on the point. The last mile between “almost good enough” and “fair enough” can be very long.
Even the simple, routine tasks that are at the heart of most job loss scenarios can be extremely difficult to automate. Amazon uses hundreds of thousands of advanced robots in its warehouses. But these aren’t androids picking up items from shelves. The robots are the shelves, which pass to humans, who still do the picking.
These simple and deft movements of the eye and hand, recognizing and grasping a myriad of three-dimensional shapes, are the product of millions of years of evolution. Scientists and engineers are working hard to catch up. But they won’t completely solve these problems all at once, not in the next nine years, nor for a long time after that.